Librarian

A corpus analysis agent combining formal information science methods with LLM judgment.

Vision

The Librarian Agent orchestrates the formal tools of information science to support information architects. LLMs are good at intuiting semantic relationships but do so as a black box. Information science has decades of rigorous, reproducible, auditable methods.

Combining LLM judgment with formal IS tools produces something better than either alone: auditable intermediate representations at each analytical stage.

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                     LIBRARIAN AGENT                             │
│  Orchestrates skills, manages deliberation, interfaces with IA  │
└─────────────────────────────────────────────────────────────────┘
                              │
         ┌────────────────────┼────────────────────┐
         ▼                    ▼                    ▼
   ┌──────────┐        ┌──────────┐        ┌──────────┐
   │INGESTION │        │  TOPIC   │        │CLUSTERING│
   │          │        │ MODELING │        │          │
   └──────────┘        └──────────┘        └──────────┘
         │                    │                    │
         ▼                    ▼                    ▼
   ┌──────────┐        ┌──────────┐        ┌──────────┐
   │VOCABULARY│        │ TAXONOMY │        │  GRAPH   │
   └──────────┘        └──────────┘        └──────────┘

What's Here

Skills (skill-librarian/)

Agent skills encoding professional practice knowledge:

• Ingestion Skill — Receive, examine, characterize a corpus. Produces a Collection Record.
• Topic Modeling (planned)
• Clustering (planned)
• Vocabulary Analysis (planned)
• Taxonomy Builder (planned)
• Concept Graph (planned)

Toolkit (skill/)

Method references and Python environment setup for corpus analysis.

TUG Blog Corpus (tug-blog/)

Example corpus: 40 articles from The Understanding Group blog.

• dashboard/ — Interactive visualization at https://librarian-murex.vercel.app
• raw/ — Scraped article text
• Analysis scripts

Key Concepts

Collection Record

The output of the Ingestion Skill. A structured description of:

• What's in the corpus
• Its statistical profile
• Prior organization detected
• Quality issues
• Feasibility assessments for different methods
• Recommendations for next steps

See skill-librarian/ingestion/references/collection-record-schema.json

Three-Phase Workflow

1. Reference Interview — Light-touch questions to understand purpose
2. Capabilities Briefing — What's possible given this corpus
3. Collaborative Triage — Options with tradeoffs, co-designed approach

Theoretical Foundation

Grounded in Glushko's Discipline of Organizing framework. See skill-librarian/references/planning-doc.md.

Local Development

# Create Python environment
python3 -m venv .venv
source .venv/bin/activate
pip install -r skill/requirements.txt

# Run dashboard locally
cd tug-blog/dashboard
python -m http.server 8888

Dashboard

https://librarian-murex.vercel.app

• Main view: Articles by persona with gap analysis
• Graph view: Topic clusters visualization